PeopleofSuperintelligence

Researcher

Jae Sung Park

Also known as James Park

Jae Sung Park is a research scientist at the Allen Institute for AI (Ai2) who works on open vision-language models that can point to and track what they see. As a PhD student at the University of Washington, advised by Ali Farhadi, Yejin Choi and Ranjay Krishna, he introduced VisualCOMET in 2020, a framework for commonsense reasoning about what happens before and after a still image. He coauthored Molmo in 2024, an open family of vision-language models built on the PixMo datasets, and the paper received a Best Paper Honorable Mention at CVPR 2025. In 2025, he was also first author of Synthetic Visual Genome, a synthetic scene-graph dataset with 5.6 million relationships. After defending his PhD in December 2025, he joined Ai2 in January 2026, and Molmo2, where he was a co-first author and led development of tracking, became a CVPR 2026 Best Paper Award Candidate.

Timeline

  1. 2019

    Adversarial video description at CVPR

    With Marcus Rohrbach, Trevor Darrell and Anna Rohrbach, he publishes a CVPR 2019 oral paper that applies adversarial techniques during inference to write better multi-sentence video descriptions. He also co-organizes the 2019 Large Scale Movie Description Challenge and the ICCV19 3rd Workshop on Closing the Loop between Vision & Language.

  2. 2020

    VisualCOMET at ECCV

    In 2020, with Chandra Bhagavatula, Roozbeh Mottaghi, Ali Farhadi and Yejin Choi, he introduces VisualCOMET, a framework for predicting what happened before a still image, what happens next and the intents of the people in it. Its Visual Commonsense Graphs hold over 1.4 million textual inferences on 60,000 images, and the paper is an ECCV 2020 spotlight.

  3. 2021–2023

    Contrast sets for video-text models

    He coauthors MERLOT, a family of multimodal neural script knowledge models, at NeurIPS 2021. As first author of a NAACL 2022 paper with Sheng Shen, Ali Farhadi, Trevor Darrell, Yejin Choi and Anna Rohrbach, he probes video-text models with automatically built contrast sets of hard negatives. In 2023, he is first author of Localized Symbolic Knowledge Distillation for Visual Commonsense Models at NeurIPS.

  4. September 2024

    Molmo and PixMo

    In September 2024, he and his coauthors release Molmo, a family of open vision-language models. Its PixMo datasets, including detailed image captions and 2D pointing data, are collected without external vision-language models, and the 72B Molmo model outperforms Claude 3.5 Sonnet and Gemini 1.5 Pro.

  5. 2025

    Synthetic Visual Genome at CVPR 2025

    At CVPR 2025, he is first author of Synthetic Visual Genome, which builds SVG, a synthetic scene-graph dataset of 146K images and 5.6M relationships, and trains the ROBIN model to construct dense scene graphs. The Molmo and PixMo paper receives a Best Paper Honorable Mention at CVPR 2025.

  6. December 2025

    Molmo2 and a UW PhD

    In December 2025, he defends his PhD in Computer Science & Engineering at the University of Washington, advised by Ali Farhadi, Yejin Choi and Ranjay Krishna. Molmo2, a family of open image and video models for pointing and tracking, is released in December 2025; he is a co-first author and leads development of its tracking capability.

  7. Since January 2026

    Research scientist at Ai2

    He joins the Allen Institute for AI (Ai2) as a research scientist in January 2026. In March 2026, he releases MolmoPoint with Christopher Clark, Yue Yang and colleagues, a pointing method that selects the visual tokens containing the target concept, and it is accepted to ECCV 2026. Molmo2 is a Best Paper Award Candidate at CVPR 2026.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Jae Sung Park yet. Send us a free-licensed photo.

  1. Jae Sung Park homes.cs.washington.edu. Checked 19 October 2026.
  2. [1812.05634] Adversarial Inference for Multi-Sentence Video Description arxiv.org. Checked 19 October 2026.
  3. [2004.10796] VisualCOMET: Reasoning about the Dynamic Context of a Still Image arxiv.org. Checked 19 October 2026.
  4. Exposing the Limits of Video-Text Models through Contrast Sets - ACL Anthology aclanthology.org. Checked 19 October 2026.
  5. [2409.17146] Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models arxiv.org. Checked 19 October 2026.
  6. CVPR 2025 Best Papers and Best Demos cvpr.thecvf.com. Checked 19 October 2026.
  7. [2506.07643] Synthetic Visual Genome arxiv.org. Checked 19 October 2026.
  8. Synthetic Visual Genome synthetic-visual-genome.github.io. Checked 19 October 2026.
  9. [2601.10611] Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding arxiv.org. Checked 19 October 2026.
  10. [2603.28069] MolmoPoint: Better Pointing for VLMs with Grounding Tokens arxiv.org. Checked 19 October 2026.